[THIS DATASET IS COMPLETELY UNUSABLE FOR AUTOMATED DATA ANALYSIS/EXTRACTION. Empty columns were filled with the next available column, somehow, causing severe inconsistencies with which columns are which. You will have to infer.]
A CSV dataset containing the metadata of roughly 149.4 million YouTube videos compiled from a January–June 2012 web crawl conducted by Common Crawl. (https://data.commoncrawl.org/cc-index/collections/index.html)
The .gz archive expands to 28 GB.
Some notes:
1. Column names are at the top of the file.
2. I suggest using Glogg to read the file; Notepad++ is horrifically slow. Import it into an SQL database if you're gonna use it often.
3. The original page can be accessed at https://data.commoncrawl.org/parse-output/segment/(SOURCE).arc.gz
3a. Keep in mind the .gz files in question are ~100mb concatenations of hundreds of compressed .html files—the beginning and ending bytes of the intended pages are (mostly) specified.
4. The source for each video is from the last time said video appeared. For videos with descriptions, their full respective pages can be accessed with their YouTube URL under "CC-MAIN-2012" on https://index.commoncrawl.org/.
4a. The pages from this 2012 web crawl are NOT on the Wayback Machine as of June 2025.
This dataset covers 95% of the YouTube pages crawled; unfortunately I've lacked the storage to complete it. Luckily, due to the massive size of the crawl (~50 million pages), very few new videos will appear by the time it's completed.